Manual Backport of Fix issue where terminating gateway service resolvers weren't properly cleaned up into release/1.13.x

[andrewstucki]

Backport

Manual backport from #16498 to release/1.13.x.

The below text is copied from the body of the original PR.

---

Description

This fixes an issue where terminating gateways weren't properly cleaning up service resolvers attached to their external services. When a resolver was deleted, the casting guard was keeping around the old ServiceResolver value due to a failed nil cast.

However, there are some questions that I do have about whether or not ServiceResolver subsets are ever supposed to work with an external service accessed through a terminating gateway, as they currently do not -- when a local proxy attempts to resolve the endpoint for an upstream that is an external service that has a ServiceResolver with subsets it winds up attempting to watch the subset upstream directly which is not returning any address (verses without a subset, returning the terminating gateway address). Either way though, that would be a different bug.

Testing & Reproduction steps

Run the following bash script:

#!/bin/bash

cleanup() {
  echo "shutting down upstreams"
}

trap 'trap " " SIGTERM; kill 0; wait; cleanup' SIGINT SIGTERM

cat << EOF | ./consul config write -
Kind      = "proxy-defaults"
Name      = "global"
Config {
  protocol = "http"
}
EOF

echo "Writing terminating gateway config entry"
cat << EOF | ./consul config write -
Kind = "terminating-gateway"
Name = "gateway"

Services = [
  {
    Name = "external"
  }
]
EOF

cat << EOF > /tmp/external.json
{
  "Node": "hashicorp",
  "Address": "127.0.0.1",
  "NodeMeta": {
    "external-node": "true",
    "external-probe": "true"
  },
  "Service": {
    "ID": "external-v1",
    "Service": "external",
    "Port": 9877,
    "Meta": {
      "version": "v1"
    }
  }
}
EOF
curl --request PUT --data @/tmp/external.json localhost:8500/v1/catalog/register

cat << EOF > /tmp/external.json
{
  "Node": "hashicorp",
  "Address": "127.0.0.1",
  "NodeMeta": {
    "external-node": "true",
    "external-probe": "true"
  },
  "Service": {
    "ID": "external-v2",
    "Service": "external",
    "Port": 9878,
    "Meta": {
      "version": "v2"
    }
  }
}
EOF
curl --request PUT --data @/tmp/external.json localhost:8500/v1/catalog/register

echo "Writing resolver config entry"
cat << EOF | ./consul config write -
Kind = "service-resolver"
Name = "external"
DefaultSubset = "v1"
Subsets = {  
  v1 = {    
    Filter = "Service.Meta.version == v1"  
  } 
  v2 = {    
    Filter = "Service.Meta.version == v2"  
  }
}
EOF

echo "Running terminating gateway"
./consul connect envoy -gateway terminating -register -service gateway -proxy-id gateway -- -l trace  &

wait

Before the fix:

➜  consul git:(terminating-gateway-resolvers) ✗ curl -s http://localhost:19000/clusters | grep external | cut -d':' -f1 | uniq | sort
external.default.dc1.internal.4e38b938-3a3e-7b67-6624-a8dae1058918.consul
v1.external.default.dc1.internal.4e38b938-3a3e-7b67-6624-a8dae1058918.consul
v2.external.default.dc1.internal.4e38b938-3a3e-7b67-6624-a8dae1058918.consul
➜  consul git:(terminating-gateway-resolvers) ✗ ./consul config delete -kind service-resolver -name external
Config entry deleted: service-resolver/external
➜  consul git:(terminating-gateway-resolvers) ✗ curl -s http://localhost:19000/clusters | grep external | cut -d':' -f1 | uniq | sort
external.default.dc1.internal.4e38b938-3a3e-7b67-6624-a8dae1058918.consul
v1.external.default.dc1.internal.4e38b938-3a3e-7b67-6624-a8dae1058918.consul
v2.external.default.dc1.internal.4e38b938-3a3e-7b67-6624-a8dae1058918.consul

After the fix:

➜  consul git:(terminating-gateway-resolvers) ✗ curl -s http://localhost:19000/clusters | grep external | cut -d':' -f1 | uniq | sort
external.default.dc1.internal.d27ec462-6e9d-edfa-bf3a-8d603e174492.consul
v1.external.default.dc1.internal.d27ec462-6e9d-edfa-bf3a-8d603e174492.consul
v2.external.default.dc1.internal.d27ec462-6e9d-edfa-bf3a-8d603e174492.consul
➜  consul git:(terminating-gateway-resolvers) ✗ ./consul config delete -kind service-resolver -name external
Config entry deleted: service-resolver/external
➜  consul git:(terminating-gateway-resolvers) ✗ curl -s http://localhost:19000/clusters | grep external | cut -d':' -f1 | uniq | sort
external.default.dc1.internal.d27ec462-6e9d-edfa-bf3a-8d603e174492.consul

PR Checklist

• updated test coverage
• external facing docs updated
• not a security concern

---

Overview of commits

• e14b430 - 5255013 - b1b2abc - ecaeff2

[nathancoleman]

Note: Diff is slightly different because the assert_upstream_missing helper didn't exist in the base branch here